Recent developments in the field of computer vision have led to advancement of various methods and techniques for visual tracking of objects by use of an imaging device, such as a camera. Currently, various techniques, either standalone or in combination, are utilized for tracking of such objects in video content. For example, template tracking technique may be utilized to track the objects, associated with rigid deformations, based on extraction of spatially sensitive features, such as histogram of oriented gradients (HOG). However, in cases where objects are associated with non-rigid deformations, the template tracking technique fails to capture or keep track of such objects, and hence provides inaccurate tracking results as the tracking gets drifted away from the original location of the objects.
In accordance with another example, histogram tracking technique utilizes a color histogram to represent the objects to be tracked. The color histogram represents spatially insensitive features and maintains the representation of the objects even if they are associated with deformations. However, the histogram tracking technique loses or discards substantial useful information, specifically, the spatial information, and hence provides inaccurate tracking results as the tracker gets drifted away to nearby regions with similar appearance.
In accordance with another example, a combination of the histogram tracking technique and the template tracking technique, in accordance with various approaches, may be utilized for tracking of the objects in the video content. In accordance with a first approach, the color histogram, generated by the histogram tracking technique, may be augmented with the spatial information that may be lost during the generation of a purely histogram based feature. Such spatial information may be augmented by the template tracking technique, however, being histogram-based, such technique remains computationally inefficient. In accordance with a second approach, patch-based histograms of the objects may be utilized to incorporate the lost spatial information. However, such approach is quite sensitive to the creation of patches in the object to be tracked and requires substantial memory for storage of multiple histograms. In accordance with a third approach, the histogram tracking technique may be fused with the template tracking technique to provide accurate tracking results. However, it may require both the tracking techniques to execute simultaneously and thus, may increase the computational complexity. Thus, an advanced, computationally efficient, and robust video processing system may be required for visual object tracking.
Further limitations and disadvantages of conventional and traditional approaches will become apparent to one of skill in the art, through comparison of described systems with some aspects of the present disclosure, as set forth in the remainder of the present application and with reference to the drawings.